Update generative-proof-of-concept-CPU-preprocessing-in-memory.py

Extend metadata for the generated text samples being logged, Add the perplexity metric and max_new_tokens.

Author: david-thrower

generative-proof-of-concept-CPU-preprocessing-in-memory.py

@@ -1089,7 +1089,7 @@ def complete_text_beam(text: str,
         # print(f"I ask the generator (Beam defaults - max_new_tokens: 10,  temperature: 0.75, top_k: 75, top_p: 0.98, repetition_penalty: None, presence_penalty: 1.3, frequency_penalty: 1.4): {test_text_block}... It responds: '{response}'.")
 
         trial_number = int(trial.number)
-        def test_text(test_prompt: str, max_new_tokens: int, sample_number: int, result: float, result_cutoff: float, trial_id: int, test_sample_number: int) -> None:
+        def test_text(test_prompt: str, max_new_tokens: int, sample_number: int, result: float, result_cutoff: float, trial_id: int, test_sample_number: int, result_0: float) -> None:
             """
             If the result < result_cutoff, this will run a matrix of different sampling values and print out the resulting text for human subjective evaluation.
 
@@ -1177,7 +1177,7 @@ def test_text(test_prompt: str, max_new_tokens: int, sample_number: int, result:
                                                 repetition_penalty=perm_0['repetition_penalty'],
                                                 presence_penalty=perm_0['presence_penalty'],
                                                 frequency_penalty=perm_0['frequency_penalty'])
-                    print(f"Trial #: {trial_id} Text Sample #: {test_sample_number} GENERATE PARAMS: temperature={perm_0['temperature']}, top_k={perm_0['top_k']}, top_p={perm_0['top_p']}, repetition_penalty={perm_0['repetition_penalty']} presence_penalty={perm_0['presence_penalty']} frequency_penalty{perm_0['frequency_penalty']} PROMPT: {test_prompt} RESPONSE: {response_0}")
+                    print(f"Trial #: {trial_id} Text Sample #: {test_sample_number} Perplexity: {result_0} GENERATE PARAMS: {perm_0['max_new_tokens']} temperature={perm_0['temperature']}, top_k={perm_0['top_k']}, top_p={perm_0['top_p']}, repetition_penalty={perm_0['repetition_penalty']} presence_penalty={perm_0['presence_penalty']} frequency_penalty{perm_0['frequency_penalty']} PROMPT: {test_prompt} RESPONSE: {response_0}")
                 #
                 # print(f"Sample {sample_number}: I ask the generator (Beam: - max_new_tokens: 10, temperature=0.6, top_k=75, top_p=0.98, repetition_penalty=None, presence_penalty = 1.3, frequency_penalty = 1.4): {test_prompt}... It responds: '{response_3}'.")
                 # response_4 = complete_text_beam(text=test_prompt, max_new_tokens=max_new_tokens, temperature=0.7, top_k=75, top_p=0.98, repetition_penalty=None, presence_penalty = 1.3, frequency_penalty = 1.4)
@@ -1207,7 +1207,8 @@ def test_text(test_prompt: str, max_new_tokens: int, sample_number: int, result:
                    result=result,
                    result_cutoff=RESULT_CUTOFF,
                    trial_id=trial_number,
-                   test_sample_number=counter)
+                   test_sample_number=counter,
+                   result_0=result)
             counter += 1
 
             # # Tokenize the text without padding first to get actual tokens